Restructuring, Pruning, and Adjustment of Deep Models for Parallel Distributed Inference